Near‐infrared spectroscopy as an alternative tool for predicting soil erodibility in a watershed under desertification

Abstract

Near‐infrared spectroscopy (NIR) has become an alternative to conventional techniques for assessing soil properties that are difficult to measure, especially at sites under intense erosion. Thus, the aim of this study was to assess near‐infrared spectroscopy as an alternative tool for predicting soil erodibility in a watershed under desertification. Two hundred and fourteen composite soil samples were collected, where 100 samples were analyzed for physical and chemical properties to obtain K (soil erodibility), Ki (interrill erodibility), and Kr (rill erodibility) values and the remaining 114 composite soil samples were only used to obtain spectral signatures in the NIR range. Then, such database was used to predict the erodibility parameters and soil physicochemical properties. Partial least squares regression was used for modeling the K, Ki, Kr indices and other soil properties. The highest erodibility indices were observed in Acrisols and Luvisols (K = 33.59 × 10−3 t h MJ−1 mm−1, R2 = 0.61, square root of the mean error (RMSE) = 2.62 and residual prediction deviation (RPD) = 1.62; Ki = 5.45 × 10−6 kg s m−4, R2 = 0.49, RMSE = 0.21 and RPD = 1.47; Kr = 21.78 × 10−3 s m−1, R2 = 0.54, RMSE = 0.66 and RPD = 1.68). Erodibility prediction using spectral data was found to be an acceptable alternative and can be applied in other areas struggling with the same phenomena to acquire new robust databases and to improve the understanding of soil erosion and desertification process. Also, our results can extend studies with universal soil‐loss equation and revised universal soil‐loss equation, since erodibility is one of the most important and expensive parameters of these equations, mainly in developing countries where financial support is scarce.